playground4rocker

In this repo there are multiple github actions and docker deployment workflows available.

I tested to run R code using different approaches. The output files are uploaded as an artifact zip of the executed jobs.

Github Actions

Results:

• With .github/workflows/r2u.yml, last R package versions with r2u are used. ETA 49s.

• With .github/workflows/pak.yml, last R package versions with pak are used. ETA 1m 18s.

• With .github/workflows/conda.yml, last R package versions with conda are used. ETA 1m 25s.

• With .github/workflows/renv.yml, locked R package versions with renv are used. ETA 12m 24s (slower but more reproducible).

Conclusions:

• When using Ubuntu, r2u is the fastest approach to run the code with the last R package versions.
• When not using Ubuntu, pak or conda approaches can be used to run the code with the last R package versions.
• When requiring specific versions, renv approach can be used. It will be slower but more reproducible.

How to build images and run containers locally

First clone the repository in your computer:

$ git clone [email protected]:jrosell/playground4rocker.git

$ cd playground4rocker

Then run the R code with the fastest approach this way:

$ docker build -f r2u.Dockerfile -t my-r2u-name . \
  && docker run --name my_r2u_container --rm  -v $(pwd)/output:/workspace/output/:rw  my-r2u-name

Alternatively, you can run R code with the more reproducible but slow approach this way:

$ docker build -f renv.Dockerfile -t my-renv-name .  \
  && docker run --name my_renv_container --rm -v $(pwd)/output:/workspace/output/:rw my-renv-name

If you have added Python code and need to run both R and Python code, you can use this version:

$ docker build -f conda.Dockerfile -t my-conda-name . \
  && docker run --name my_conda_container --rm -v $(pwd)/output:/workspace/output/:rw my-conda-name

Follow up

Feedback is welcome:

• Suggestions? Bugs? You can open an issue.
• You can fork this repo an reuse it. I'm open to PR too.